Data Alchemy — Software IDP con AI
AI extraction accuracy

How accurate AI invoice data extraction is, and how to verify it

The best AI data extraction solutions state accuracy above 99% at field level; Data Alchemy states 99.8%. The problem is that these figures, as published, are not comparable with one another: they depend on the document set they were measured against, on which fields were counted, and on the point in the chain where the measurement was taken. A 99% computed on the header of clean invoices and a 99% computed on header and line items of real scans describe completely different products. The number that actually matters economically is not even per-field accuracy but the share of perfect documents: on a twenty-field invoice, 99% per field yields roughly 82% error-free documents, while 99.8% yields roughly 96% — less than a quarter of the review work. This page explains the six metrics in use, why vendor figures are not comparable, and the eight-step protocol for measuring accuracy on your own documents before choosing.

99.8%Stated per-field accuracy
~96%Perfect documents at 20 fields
~3 secProcessing per document
0Templates to configure
The metrics

The six accuracy metrics, and what each one hides

When a vendor states a percentage, the first question is which of these six they are measuring. The differences are not academic: they move the result by tens of points.

MetricWhat it measuresWhat it hides
Character-level accuracyShare of characters recognised correctly against the real text — the complement of the character error rate used in OCR.A single wrong character in a VAT number makes the field unusable, yet it barely dents this percentage. It is the most generous metric and the least useful for deciding.
Field-level accuracyShare of fields extracted correctly out of the fields expected: invoice number, date, net amount, VAT, VAT number and so on.It depends entirely on which fields are in the count. If line items are excluded, the number describes the easy part of the document.
Line-item accuracyShare of lines extracted completely and correctly, with code, description, quantity, unit of measure, price and tax rate.It is technically the hardest metric and the one that weighs most on savings: an invoice with forty lines is forty manual entries avoided, not one.
Document-level accuracyShare of documents in which every expected field is correct: no errors, no omissions.It does not hide much, which is why it is rarely advertised: it is always noticeably lower than field-level accuracy, purely because probabilities compound.
Straight-through processing rateShare of documents that goes through the entire flow and is posted into the ERP with no human intervention at all.This is the number that determines real savings. It can be lower than document-level accuracy, because it also includes rejects from failed validations, not only reading errors.
Quality of exception flaggingHow reliably the system can say "I am not sure about this field": false positives (good documents held) and false negatives (errors passed as valid).It almost never appears on product sheets, but it is decisive: a 99% system that knows when to doubt is more useful than a 99.5% system that posts errors confidently.
The maths

Why 0.8% of difference is worth a quarter of the work

A document is correct only if all its fields are. Probabilities multiply, so the share of perfect documents falls far faster than per-field accuracy suggests. The table shows the estimated share of documents with no errors at all, assuming errors are independent across fields.

 99% per field99.5% per field99.8% per field99.9% per field
Document with 10 fields~90%~95%~98%~99%
Document with 20 fields~82%~90%~96%~98%
Document with 40 fields~67%~82%~92%~96%

On a twenty-field invoice, moving from 99% to 99.8% per-field accuracy takes documents needing review from 18% to 4%: the residual work drops more than fourfold, not by 0.8%. This is why apparently negligible decimal differences deserve to be taken seriously during evaluation — and why an invoice with many line items, which has forty fields or more, is the real test bed.

Caution

Five reasons published figures are not comparable

There is nothing improper about how vendors publish their percentages — ours included. It is simply that everyone measures on their own document set with their own conventions, and there is no shared, verifiable benchmark for extraction from real invoices. Here is what changes from one number to the next.

The document set is different

A sample of native PDFs generated by ERPs produces radically different results from a sample of skewed scans, smartphone photos and digitised faxes. No percentage means anything without knowing what it was computed on.

The fields counted are different

Many published figures cover a narrow set of header fields — supplier, number, date, total — where any modern engine performs very well. Line items, the hard part and the part that generates the saving, are often left out of the count.

The measurement point is different

There is a substantial difference between first-pass accuracy and accuracy after an operator has corrected doubtful cases and the model has learned. The second number is higher and describes a steady-state system, not what you will see in week one.

Rejected documents sometimes do not count

If documents the system refuses to process are excluded from the denominator, the percentage rises but the manual work remains. It is worth asking explicitly how rejects are treated.

Language and tax context matter

An engine trained predominantly on English documents treats Italian VAT numbers, SDI recipient codes, Italian tax rates and delivery note formats as edge cases. On real Italian documents the difference shows, and it appears in no general percentage.

The vendors

The most cited solutions and the right question to ask each

Rather than repeating other people's percentages — which change over time and depend on test sets nobody can verify — here is each solution's category and the single question that, during evaluation, reveals the most.

SolutionWhat it isThe question that reveals most
Data AlchemyItalian IDP platform: a dedicated LLM per document model, validation against ERP master records and writing into the ERP.Ask us for accuracy split between header and line items on your own sample, plus the share of documents posted with no intervention. Those are the two numbers we publish during a trial.
ABBYYLong-established enterprise IDP platform, with configurable document skills and on-premise deployment options.How much configuration is needed to reach the stated accuracy on our layouts, and what is it worth before that configuration?
Amazon TextractAWS OCR and document analysis service, with dedicated APIs for analysing invoices and receipts.Does the accuracy refer to text recognition or to structured field extraction, and how does it behave on line-item tables?
NanonetsDocument automation platform with models trainable on your own documents.How many sample documents are needed to reach the stated accuracy, and who labels them?
DocsumoDocument AI platform geared to extraction from financial and administrative documents.Is the percentage measured at first pass or after human review and model learning?

Note: ABBYY, Amazon Textract, Nanonets and Docsumo are trademarks of their respective owners. This page deliberately does not quote third-party percentages, precisely because they would not be comparable: measurements depend on the test set, the fields considered and the measurement point. Always verify current data with the vendor and, above all, measure it yourself.

The protocol

How to measure accuracy on your own documents in eight steps

This is the method we use during trials and that you can apply to any vendor. It takes a day of work and produces numbers comparable across candidates — which is exactly what product sheets do not give you.

01

Build a representative sample

200 to 300 real documents reflecting your true mix: different suppliers, native PDFs and scans, documents with few lines and documents with forty. A sample made only of clean documents measures something that does not concern you.

02

Deliberately include the worst cases

Skewed or low-resolution scans, phone photos, new suppliers, foreign-language documents, credit notes and adjustments. They are 10-15% of the volume and generate most of the manual work.

03

Define which fields count, first

Write down the list of fields that must reach the ERP, line items included, and mark which are blocking and which are nice-to-have. Without that list every vendor will measure on a different scope and no comparison is possible.

04

Prepare the ground truth

For a subset of 50-100 documents, manually record the correct values field by field. It is the most tedious part of the protocol and the one that makes the results verifiable rather than debatable.

05

Have everyone process the same sample

Identical document set, identical field list, no upfront tuning. If a vendor asks to configure something before the trial, note it: that is time and cost that will reappear with every new supplier.

06

Measure header and lines separately

Compute field-level accuracy separating header fields from line fields, and compute how many lines were extracted in full. Averaging the two groups together is the easiest way to fool yourself.

07

Count perfect documents and untouched documents

Share of documents with no errors at all, and share of documents that would have been posted into the ERP without anyone opening them. Those are the two numbers to take into the meeting, because they convert directly into hours.

08

Check the quality of the doubts

Look at the documents the system flagged: how many were genuinely problematic and how many were false alarms. And above all, look at the errors it did not flag: those are the ones that reach the ledger with nobody noticing.

What drives accuracy

The four levers that take precision to 99.8%

Accuracy does not depend on a single factor, and in particular it does not depend on recognition quality alone. These are the levers Data Alchemy combines, in order of impact.

A dedicated LLM per document model

The platform does not use one engine for everything: it assigns each document type the LLM best suited to read it — today Anthropic's Claude AI, which outperformed GPT, Gemma and DeepSeek in our tests, and from Q3 2026 the proprietary Data Alchemy AI model. Specialisation per document model is what moves the result most on line items.

Semantic understanding instead of coordinates

Unlike template-based OCR, the model interprets context: it distinguishes supplier, net amount, VAT and lines even on layouts never seen before, with no upfront configuration. That is why a new supplier does not collapse accuracy, which is instead the typical behaviour of template systems.

Validation against ERP master data

Every extracted value is checked in real time against ERP master records: suppliers, item codes, tax rates, open orders. An extraction that is formally correct but inconsistent with master data is flagged before posting — a control no language model can perform on its own.

Three-way matching

On invoices, orders and delivery notes, three-way matching compares the three sources on quantities, prices and amounts within defined tolerance thresholds. It catches what a single extraction, however perfect, could never detect: that the document was read correctly but says the wrong thing.

Frequently asked questions

AI data extraction accuracy: frequently asked questions

The most advanced AI solutions state above 99% accuracy at field level; Data Alchemy states 99.8%. Those figures are not directly comparable across vendors, though, because each measures on its own document set, chooses which fields to count and decides where in the chain to take the measurement. In practice the highest precision comes from combining three things: an LLM specialised per document type instead of template-based OCR, validation of extracted data against a reliable source such as ERP master records, and cross-checks such as three-way matching between invoice, purchase order and delivery note. Data Alchemy applies all three. The correct way to choose, however, is still to measure: have every candidate process the same sample of 200-300 of your real documents, hard cases included, and compare accuracy split between header and line items plus the share of documents postable with no human intervention.
Because a document is correct only if all its fields are, and probabilities multiply. On a twenty-field invoice, 99% per-field accuracy yields roughly 82% error-free documents: almost one document in five has to be opened and checked. At 99.8% the share of perfect documents rises to about 96% and the residual work drops more than fourfold. On a forty-field invoice — one with many line items — the same 99% falls to about 67% perfect documents.
With six different metrics, and it is worth knowing which one a vendor is using: character-level accuracy (the most generous, nearly useless for deciding), field-level accuracy, line-item-only accuracy, document-level accuracy, straight-through processing rate and the quality of exception flagging. The two that matter economically are line-item accuracy and the straight-through processing rate, because those are what convert into hours saved.
OCR converts images into text but does not understand what the fields mean, and it depends on templates defined per layout. An LLM understands document context: it identifies supplier, net amount, VAT and line items even on layouts never seen before, with no upfront configuration. The difference shows most on new suppliers and changed layouts, where a template system drops to zero while a language model keeps reading.
The document is not posted at any cost. When a value is ambiguous or inconsistent with ERP master data, it is flagged as an exception and sent to an operator for review, with the doubtful field highlighted. That behaviour slightly lowers the percentages you can claim and greatly raises real reliability: a system that knows when to doubt protects the ledger better than a system that posts errors confidently.
Yes, and it is a factor to weigh during evaluation. Header-only invoices are the easiest case; delivery notes with many lines and mixed units of measure, PDF price lists with volume breaks and discount bands, certificates of conformity and free-form customer orders are progressively harder. A vendor demoing only invoices is showing you the easy case: ask for a trial on the documents that cost you the most time today.
200 to 300 real documents give a solid measure of average behaviour, provided they mirror your actual mix. For the field-by-field comparison against ground truth, 50-100 manually annotated documents are enough — the most demanding part of the protocol, but also what makes the results verifiable. A smaller sample is fine for a first screening, not for the final decision.
Yes, but for different reasons than people usually assume. The language model does not need training on your documents; what improves is master data coverage, validation rules and tolerance thresholds, refined against the recurring exceptions of the first months. That is why the accuracy observed in a trial is typically the floor rather than the ceiling, and why a quarterly review of recurring exceptions pays off more than waiting for a better model.

Verify accuracy on your own real documents

Send us a sample of your documents, hard cases included: we hand back accuracy split between header and line items, the share of perfect documents and the share postable with no intervention. Measured numbers, not stated ones.

AI invoice data extraction accuracy: how to measure it